Session Item

Dysphagia is a common consequence of head and neck (HN) RT, which negatively affects quality of life. Voxel-wise image-based data mining (IBDM) allows exploring the relation between RT dose and treatment outcome without prior assumptions. Here, we used IBDM to explore the relation between RT dose to normal HN anatomy and dysphagia at 1 year after treatment, assessed using three validated outcome measures: composite M.D. Anderson Dysphagia Inventory (MDADI), performance status scale for normalcy of diet (HN-PSS), and water test (WT).

IBDM was applied to data from 104 oropharyngeal cancer patients treated with definitive (Chemo)RT and assessed for swallow function pre-treatment and at 1 year by a Swallow Assessment Specialist. We spatially normalised all patient’s planning dose matrices (in equivalent dose at 2Gy/fraction, α/β=3Gy) to three arbitrary reference anatomies using a combination of thin-plate splines deformation (for neck alignment) and NiftyReg. To reduce target laterality bias, all patients were mirrored in the left-right axis and mapped twice to the reference patient. Regions where the dose was associated with outcome at 1 year were found by performing voxel-wise statistics (t-test for WT, Spearman correlation for MDADI/HN-PSS) and permutation testing (n=1000).

Mean and maximum doses of the overlap of the significant regions for all outcomes were averaged across the three reference patients. Clinical factors, treatment variables and pre-treatment measures for WT, HN-PSS and MDADI (table 1) were used in multivariable analysis to predict WT (logistic regression), HN-PSS and MDADI (linear regression) at 1 year. A clinical model was found using backward stepwise selection for each outcome. Improvement of model discrimination after adding the mean/max doses was quantified using Akaike information criterion (AIC).

IBDM revealed significant association between dose to distinct regions and the three outcomes tested (p<0.005), overlapping around the inferior section of the brainstem (figure 1). Models for WT and HN-PSS at 1 year were significantly improved by including mean dose to the overlap region (p=0.02); this was not the case for the MDADI model. Inclusion of max dose did not significantly improve discrimination of any model.

Mean dose to the inferior section of the brainstem is strongly associated with dysphagia at 1 year following HN RT. The identified region is in proximity to swallowing centres in the medulla oblongata, providing a possible mechanistic explanation. Unexpectedly, no significant region was found close to the pharyngeal constrictor muscles or salivary glands, despite dose to these structures being predictive for dysphagia in previous studies. Mean, rather than max dose, was most predictive, possibly indicating a parallel organ effect or a serial effect in small substructures (masked by imperfect registration). Research is required to derive a dose constraint for RT planning.